CSC 2231 review of Pick Two

This paper illustrates the main obstacle for providing a highly available
p2p storage system is the maintenance overhead of replica under p2p churn.
It uses a simple data maintenance model to derive the trend between the
number of hosts and the average membership time under cable modem or
dial-up bandwidth. It shows it is hard to achieve high availability in a
dynamic p2p storage networks under current user bandwidth conditions.

However, the paper targets at a very high availability, i.e. six nine, in
their analysis. Maybe even with one nine availability, the users of p2p
storage system are satisfied since they get it for free. So, the question
here is that what a reasonable average availability is for a general p2p
storage system.

Moreover, this paper does not differentiate the characteristics of files.
Files in p2p storage may have diverse popularity, and thus we should
use different replication strategies for them so that popular files have
higher availability and unpopular ones have lower availability. Maybe we
also need distinguish large files and small files.

This paper mentioned incentive users. I am thinking whether it is possible
to follow the mode of planet lab. If people want to have a stable and
large storage, the quota or the downloading rate that a peer gets is
dynamically related with the average active time of the peer. This papers
shows the benefits of erasure coding is very promising. It is unclear
whether this coding is general effective for all files.

Although its model is simple, and considers all files equally, this paper
is good to tell us the main challenge for current p2p storage system is
p2p churn instead of looking up time.
Received on Mon Nov 14 2005 - 01:50:56 EST